DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Adds a WSL keepalive toggle in DSH WebUI settings that prevents idle distribution shutdown via dbus-daemon, with live PID display and precise start/stop control.
Retries LLM-generated topic titles for every dsh session at startup and turn-end, plus a hover-card size readout.
Upload or paste images in the chat box; on send, transcribe image content via DashScope vision model or Windows offline OCR and inject the description into the message.
Turns AMDGPU HSACO or ELF files into queryable JSON exposing code-object metadata and ISA instruction streams, without executing kernels or predicting performance.
Ships Chinese translations of dsh's four stock agent presets (standard, cordis, code, minimal) and loads them by locale, reducing inference translation overhead for Chinese-speaking users.
Adds image recognition to DSH sessions via SiliconFlow vision models, supporting local files, URLs, and Base64 inputs.
Automatically populates missing reasoning-effort declarations for recognized reasoning models in OpenAI-compatible custom providers within DeepSeek Harness.
A DeepSeek Harness model-inference plugin that applies minimal rule constraints to guide the reasoning process.
Convert JSON data into TypeScript interfaces, JSON Schema, or Python Pydantic models with automatic type and nesting inference, fully local.
A pure Node.js string similarity plugin that runs locally with no network or external service dependencies.
Cromus platform plugin that adds skill scoring, spec validation, MCP/model drift detection, and cost estimation to the dsh agent via 11 built-in tools.
A universal project AI documentation protocol that maintains a per-project docs/.ai/ knowledge base, ensuring agents read the index before acting and write back changes to prevent context drift.